Unmanned aerial vehicle path planning for surveillance missions with regions of interest

Abstract

Surveillance missions entail visiting defined sites of interest in an environment to perform a desired task. The core of the decision making lies in planning the path and trajectory to achieve the goals assigned with the task; often modeled as a Traveling Salesman Problem (TSP). Several algorithms exist in literature for tackling the TSP and its variants. However, the problem is often accompanied with heavy computational effort. As a result, in several distinct approaches in the literature, the optimality of the solution comes at the expense of computational efficiency. In addition, each of the existing algorithms addresses a specific version of the problem and may not be applicable under different constraints or conditions imposed on the system. In this thesis, a path planning algorithm is proposed for surveillance missions that do not only require visiting certain known specified regions, but also exploring these regions. A planar kinematic model with bounded steering rate is set as a motion model, and the region exploration is imposed as traversing an arc with a nonzero arc-length around a point. Using the above two assumptions, a time-optimal control problem is defined to ensure completely visiting and exploring all of the sites, formulated as a Dubins Traveling Salesman Problem with Neighborhoods. This study builds on previous work done on the convex optimization solution to derive a geometrical representation of the local minima for path headings. Hence, solving for the optimal local Dubins maneuver between every pair of consecutive regions along the visiting sequence that corresponds to the global shortest path. The solution is validated to converge to that of the brute force approach with simulation results demonstrating the high computational efficiency of the proposed algorithm for maps with up to 500 regions.